Model Risk Dashboard

Spot questionable job postings before they hit applicants.

Score listings using the calibrated pipeline, inspect the features that drive decisions, and keep an eye on model health metrics - all in one vertical flow.

Threshold:http://localhost:8000
Score a job postingAwaiting submission
Paste details from a listing. We keep the text local until you submit for scoring.
Metadata flagsThese booleans align with the training data and directly influence the model's tabular features.
Fields marked * are required. We automatically apply your gray-zone policy.
Fraud probability
-
Binary label
Pending
Threshold applied
-
Gray-zone band: -
  • Text goes through the same TF-IDF + tabular pipeline used during training.
  • Calibrated probabilities are compared to the current threshold and gray-zone policy.
  • Predictions return decision labels along with metadata so you can log or review them.
Performance trends
Threshold tuning and latency envelopes from the latest training benchmark.
F1 vs. thresholdLatest F1: -
Latency envelopesThroughput @ batch 32: - rps
Slices to review
Lowest F1 slices from the latest evaluation so you can focus manual audits where the model struggles most.
Model snapshot
Validation and held-out test scores from the most recent training run.
Model leaderboard
Recent training runs captured in the lightweight tracker. Default order is Test F1 (high to low).
Feature signals
Inspect how tokens influence the calibration stack across fraud (positive) and legitimate (negative) labels.